肺部電腦斷層掃描影像之肺腫瘤邊緣描繪與良惡性分類

Modeling Context in Pulmonary CT Images: Applications to Nodule Classification and Segmentation

鄒奇軒(Chi-Hsuan Tsou)

指導教授 ： 陳中明

國立臺灣大學/工學院/醫學工程學研究所/博士(2015年)

https://doi.org/10.6342/NTU.2015.00147

摘要

從不同的醫療儀器所產生的影像中，如何辨識目標物件或區域，對於醫學影像分析而言，是一個重要的問題。於先前的研究，其主要目標是專注在影像中的單一物件之描繪，例如：心臟、肺部或肝臟等器官。近期研究則以偵測或定位多個組織結構，且應用於三維電腦斷層掃描影像。但還有少數的研究，並非只偵測單一或多個物件或器官，而是將擷取和辨識技術整合於同一架構。此架構所採用的技術為納入物件周圍的資訊，且已被證實能夠提升物件辨識或偵測的效能。
於本論文，我們亦考慮物件周圍的資訊，應用於肺癌電腦輔助診斷系統中的兩項主要元素：腫瘤良惡性分類和腫瘤邊緣擷取。並且嘗試回答下列問題：第一：肺部電腦斷層掃描影像中的哪些資訊，可有效地被使用於肺癌風險評估？我們採用腫瘤與其周圍資訊，結果顯示能夠有效提升預測肺腫瘤的惡性機率。第二：如何辨識肺部電腦斷層掃描影像內的解剖構造？研究結果顯示，結合統計式區域合併與條件式隨機模型，可以透過圖形劃分最佳化，得到多項特定的肺部組織結構。

關鍵字

肺部電腦斷層掃描影像 ； 腫瘤良惡性分類 ； 毛玻璃樣腫瘤邊緣擷取 ； 統計式區域合併 ； 條件式隨機模型 ； 階層式結構樹

並列摘要

The task of recognizing every object/region in images acquired with different medical imaging modalities is a key problem in medical image analysis (also called image understanding). Most of the earlier object/organ recognition algorithm assigns a single label to an image, e.g. an image of a heart, a lung or a liver. Some go further in detecting and localizing multiple anatomical structures within three-dimensional computed tomography (CT) scans. Instead of applying to a single or multiple object/organ detection, a few concurrent approaches combine segmentation and recognition into one coherent framework. Incorporating contextual information into coherent framework has proven to enhance performance of higher level tasks such as object recognition or detection.
In this thesis, we adopt the concept of image understanding to investigate two key components of computer-aided diagnosis (CAD) system for lung cancer, namely nodule classification and nodule segmentation. Specifically, we ask two questions. First: What information from pulmonary CT images can be helpful as context for lung cancer risk prediction? We show that recognition of anatomic patterns of pulmonary nodules can be potentially useful and robust algorithm in predicting the probability of the malignancy of pulmonary nodules. Second: How can recognition of anatomic structures in pulmonary CT images be performed? We show that given an image, semantically meaningful regions each labeled with a specific lung tissue class can be extracted by unifying the techniques of statistical region merging and conditional random field (CRF) with graph cut optimization.

並列關鍵字

Lung CT images ； Nodule classification ； Ground-Glass nodule segmentation ； Statistical region merging ； Conditional random field ； Hierarchical segmentation tree

參考文獻

Criminisi A, Robertson D, Konukoglu E, Shotton J, Pathak S, White S, Siddiqui K. Regression forests for efficient anatomy detection and localization in computed tomography scans. Med Image Anal. 2013(b) Dec;17(8):1293-303.

Crouzet, S.M., and Serre, T., “What are the Visual Features Underlying Rapid Object Recognition?,” Frontiers in Psychology 2, (2011).

Dehmeshki, J., Amin, H., Valdivieso, M., Ye, X.: Segmentation of pulmonary nodules in thoracic CT scans: a region growing approach. IEEE Trans Med Imaging. 27, 467–80 (2008)

El-Baz A, Beache GM, Gimel'farb G, et al. Computer-aided diagnosis systems for lung cancer: challenges and methodologies. Int J Biomed Imaging 2013;2013:942353.

Gavrielides MA, Kinnard LM, Myers KJ, Petrick N. Noncalcified lung nodules: volumetric assessment with thoracic CT. Radiology. 2009 Apr;251(1):26-37.